Refactor thread #118
Refactor thread #118
Conversation
|
The thread code belongs in core, not rt/. The rt/ directory is for compiler support code. If we want to hide the implementation, why not make core/internal to match Phobos' std/internal? Also, why make ThreadImpl a struct instead of a class? I can see having a struct Thread wrapper that's passed around by value, but since ThreadImpl would always be dynamically allocated, why a struct? To avoid GC allocation? |
|
core/internal sounds fine Your probably right about making ThreadImpl a class, won't hurt to have two more. |
|
Any further objections? |
|
Seems like I messed up the merge. Will clean this up tomorrow. |
- It is one of the most interdependent modules in the whole runtime but at the same time it is implemented completely in the runtime interface. - Deriving Thread and Fiber is part of the specified API. By reducing the member to a single pointer we'll remain ABI compatible with implementation changes. - This adds another level of indirection for every Thread and Fiber method which is neglectable given the complexity of their methods. - Use a header file and an implementation file to avoid a pseudo namespace using dozens of extern(C) functions - Compilation of the implementation needs to be separated from header usage because the compiler currently doesn't support this.
- fix module clash of core.thread header and implementation - removed struct scope (thread_aux)
|
Bump. This is needed to merge the rest of the shared library support for ELF. If there is any disagreement with that please get back to me so that we can sort it out. |
|
Rather than pimpl, what if core/thread.di were hand-written to hide all implementation details? Is there really a need for the double indirection here? |
|
The double indirection buys you a stable ABI, which will be an increasing concern once I was initially trying to further break down the whole module but that failed. |
Not much refactoring yet but this is a requirement to
integrate threads with other internal druntime components.